Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-identification

This paper considers the domain adaptive person re-identification (re-ID) problem: learning a re-ID model from a labeled source domain and an unlabeled target domain. Conventional methods are mainly to reduce feature distribution gap between the source and target domains. However, these studies largely neglect the intra-domain variations in the target domain, which contain critical factors influencing the testing performance on the target domain. In this work, we comprehensively investigate into the intra-domain variations of the target domain and propose to generalize the re-ID model w.r.t three types of the underlying invariance, i.e., exemplar-invariance, camera-invariance and neighborhood-invariance. To achieve this goal, an exemplar memory is introduced to store features of the target domain and accommodate the three invariance properties. The memory allows us to enforce the invariance constraints over global training batch without significantly increasing computation cost. Experiment demonstrates that the three invariance properties and the proposed memory are indispensable towards an effective domain adaptation system. Results on three re-ID domains show that our domain adaptation accuracy outperforms the state of the art by a large margin. Code is available at: https://github.com/zhunzhong07/ECN

PDF Abstract CVPR 2019 PDF CVPR 2019 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Unsupervised Person Re-Identification DukeMTMC-reID ECN Rank-1 63.3 # 9
Rank-10 80.4 # 8
Rank-5 75.8 # 8
MAP 40.4 # 9
Unsupervised Person Re-Identification DukeMTMC-reID->Market-1501 ECN mAP 43 # 4
Rank-1 75.1 # 3
Rank-10 91.6 # 1
Rank-5 87.6 # 1
Rank-20 94.5 # 1
Unsupervised Person Re-Identification DukeMTMC-reID->MSMT17 ECN mAP 10.2 # 4
Rank-1 30.2 # 3
Rank-10 46.8 # 2
Rank-5 41.5 # 1
Unsupervised Domain Adaptation Duke to Market ENC mAP 43.0 # 17
rank-1 75.1 # 16
Unsupervised Domain Adaptation Duke to MSMT ECN mAP 10.2 # 9
rank-1 30.2 # 9
rank-5 41.5 # 8
rank-10 46.8 # 9
Unsupervised Person Re-Identification Market-1501 ECN Rank-1 75.1 # 15
MAP 43 # 14
Rank-10 91.6 # 11
Rank-5 87.6 # 11
Unsupervised Person Re-Identification Market-1501->DukeMTMC-reID ECN mAP 40.4 # 4
Rank-1 63.3 # 4
Rank-10 80.4 # 2
Rank-20 84.2 # 1
Rank-5 75.8 # 2
Unsupervised Person Re-Identification Market-1501->MSMT17 ECN mAP 8.5 # 4
Rank-1 25.3 # 4
Rank-10 42.1 # 2
Rank-5 36.3 # 1
Unsupervised Domain Adaptation Market to Duke ENC mAP 40.4 # 17
rank-1 63.3 # 15
Unsupervised Domain Adaptation Market to MSMT ECN mAP 8.5 # 10
rank-1 25.3 # 10
rank-5 36.3 # 9
rank-10 42.1 # 10

Methods


No methods listed for this paper. Add relevant methods here